The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs network requests to sheets.googleapis.com using curl. As this is a well-known service and the operations are necessary for the skill's stated spreadsheet management functions, this activity is considered safe.
[CREDENTIALS_UNSAFE]: The skill correctly utilizes the GOOGLE_SHEETS_TOKEN environment variable for authentication. This secret is declared in the vm0_secrets metadata field, adhering to the platform's standard security practices for credential handling.
[COMMAND_EXECUTION]: Local execution of curl and jq is used to send API requests and process JSON responses. These are standard tools for programmatic web interaction and do not involve suspicious or obfuscated parameters.
[PROMPT_INJECTION]: The skill reads data from external spreadsheets, which constitutes an ingestion point for untrusted content and thus a surface for indirect prompt injection.
Ingestion points: Spreadsheet cell data is fetched through curl commands in SKILL.md.
Boundary markers: None are explicitly used in the command templates.
Capability inventory: Network access to Google APIs and local command execution via curl and jq.
Sanitization: No content validation or sanitization is implemented within the provided templates; the skill assumes the integrity of the data returned by the API.

google-sheets